* feat(a2a): deliver latest incoming message preview into the claim briefing
list_unread_a2a now carries last_message_preview (the latest message from the
OTHER agent, never the agent's own reply), fetched via a correlated subquery in
the same query — no N+1 on the per-verb briefing path.
* feat(a2a): read_a2a verb delivers unread message bodies to the agent
A2AService.get_unread_messages returns the caller's unread INCOMING messages
(never its own sends), marking exactly those rows read atomically so a message
arriving mid-call is preserved. Wired as the read_a2a content verb (route +
do_server tool + granted to every delivery role) — the content-bearing read the
A2A inbox lacked (read_messages only zeroed the counter).
* docs(rag): document read_a2a as the A2A content-read path
* fix(task): backlog activation no longer requires a discussion session
Removes the SessionTaskTable gate in activate() (and its dangling log field),
deletes _inherit_parent_session + its create() call, and drops the now-unused
SessionTaskTable import. Coordination rides task state; the session subsystem is
being retired. Tests updated to the new (no-session) behavior.
* fix(orchestrator): drop session sweep from _run_sweep
Removes the messaging import + sweep_timed_out_sessions call. That import sat
outside the try/except, so once messaging.py is deleted it would have killed the
entire sweep cascade (budget kill-switch, token rollups, retention, image prune,
superseded-PR reconcile). Notification sweep + all maintenance sweeps unchanged.
* release-manager --no-tags read-clone fix
* test: update evidence_repo unit test for a2a last_message_preview
* refactor(gateway): drop session propagation on delegate
Removes propagate_sessions_to_subtask from delegate(), the ChoreographerDeps
messaging field + property, and the ChoreographerDeps messaging arg in deps.py
(ContentActions messaging + import stay until the verbs are removed). Deletes the
propagation test; strips the now-invalid messaging kwarg from ChoreographerDeps
test builders.
* refactor(gateway): remove say/open_session/link_session/channels verbs
Removes the four channel/session verbs across content_actions (impls +
ContentActionsDeps.messaging), do_server (tools + registry), role_config (grants
+ _CHANNEL_DISCOVERY), do.py (routes), schemas/v1/do.py (request models), and
deps.py (MessagingService import + construction). Regenerates the prompt verb
tables. dm/notify/read_messages/read_a2a stay. Tests deleted/updated accordingly.
* uv.lock Upgrade
* refactor: remove conversation RAG indexing; Secretary announces via notification
Drops the CONVERSATIONS index (index_conversation, ConversationsIndexPlugin,
IndexType.CONVERSATIONS enum, IndexConversationParams, mentor.py type-label, the
messaging index hook) and its chunk-table manifest entries. The Secretary's
ANNOUNCE/RELAY_MESSAGE now fan out a BROADCAST notification to every agent's
inbox (NotificationService.broadcast) instead of posting to a dead channel.
* fix(panel): label RAG health error lines by subsystem
A red llm_error (e.g. the glm-5.2:cloud weekly-limit 429) rendered under
the 'Embedding: ok' header with no label, reading as an embedding failure.
Prefix each error line with LLM / Embedding / Vector store.
* refactor: remove channel/message reads from metrics, dashboard, git, events
MetricsService drops get_communication_volume + the MessageTable
message-count in get_agent_metrics (and the now-dead messages_sent_week
field). DashboardService drops get_channel_feeds/_compute_channel_status
and the message read in get_recent_activity (task activity kept);
get_auditor_metrics no longer reports communication_volume.
GitService's two primary-session-id helpers always return None now
(callers already treat None as "no primary session"). events/handlers.py
drops the SESSION_CLOSED/SESSION_TIMEOUT subscriptions + the
handle_session_boundary handler.
Forced follow-on: api/routes/dashboard.py + api/schemas/dashboard.py
dropped the now-dangling live_feeds/ChannelFeed surface and the
/metrics/communication route, which wrapped the removed service calls
directly (mypy would otherwise fail on the missing attributes).
* refactor: delete MessagingService + channel seeding
Edited db/__init__.py and services/__init__.py first (drop the unconditional
Channel/Group/Message/Session table + MessagingService re-exports), then
deleted services/messaging.py, then trimmed db/seed.py to only create_agents
(create_channels/create_channel_memberships/create_initial_messages gone).
Forced expansion: api/routes/{channels,groups,sessions,messages}.py import
roboco.services.messaging directly (not through the package __init__), as
does api/routes/tasks.py (the session-links embed on GET /tasks/{id} and the
GET /{id}/sessions route). Deleting messaging.py without addressing these
breaks `import roboco.api.app` immediately, since app.py eagerly imports all
route modules at startup. Since the 4 CRUD route files are 100%
MessagingService-backed with zero independent logic (and are wholesale
deletes in the plan's later API-routes task anyway), deleted them now +
unmounted from app.py/routes/__init__.py; tasks.py got the same surgical
trim its later task already specified (drop session-links embed +
TaskSessionLinkResponse/TaskResponse.sessions). This pulls a slice of that
later work forward — the routes/schemas for channels/groups/sessions/messages
still need their own pass, but their messaging-coupled parts are gone.
Verified with a full-suite collection sweep (12010 tests collected, zero
import errors) beyond the directly touched test dirs, given the expanded
blast radius.
* refactor: remove channel/session/message models, tables, and channel policy
Models: deleted channel.py/group.py/session.py/messaging.py wholesale
(zero external consumers besides the models/__init__.py re-export).
message.py surgically trimmed: removed MessageCreate (dead) and MessageEdit
(never instantiated; ExtractedMessage.edit_history retyped to
list[dict[str, Any]] to match how it's actually persisted — confirmed
ExtractedMessage was never written to any DB table, so MessageTable's
removal carries no functional risk to the kept extraction pipeline).
base.py: removed SessionStatus + ChannelType, kept MessageType. Also
removed the confirmed-dead channels_read/channels_write fields from
models/agent.py:AgentPermissions and models/dashboard.py:ChannelFeedData.
db/tables.py: deleted ChannelTable/GroupTable/SessionTable/SessionTaskTable/
MessageTable, TaskTable.session_links, and JournalEntryTable.session_id —
cascaded through models/journal.py, services/journal.py, and
api/schemas+routes/journals.py (22 plumbing sites).
foundation/policy/communications.py: removed the ChannelSpec/CHANNELS
catalog + TEAM_SCOPED_ROLES/_CELL_*/_AUDITOR_ONLY helpers, kept the
notification policy (Priority/parse_priority/NOTIFY_SENDER_ROLES/
ACK_REQUIRED_BY_TYPE). enforcement/channel_access.py deleted (confirmed
fully dead in production). agents_config.py: removed CHANNEL_ACCESS
(kept A2A_ALLOWED_PAIRS). seeds/initial_data.py: removed
DEFAULT_CHANNELS/CHANNEL_MEMBERSHIPS/AUDITOR_SILENT_ACCESS + the
never-consumed INITIAL_MESSAGES. config.py: removed
session_idle_timeout_seconds (zero consumers). exceptions.py: removed
dead ChannelError/ChannelAccessDeniedError/SessionClosedError.
Forced expansion beyond the original file list — ChannelType cascaded
into a live, mounted surface the plan didn't trace: agents_config.
CHANNEL_ACCESS -> services/permissions.py's channel-RBAC methods (not
models/permissions.py, which turned out to have no channel code at all)
-> two real endpoints in api/routes/stream.py (GET /permissions,
GET /permissions/channel/{name}) and two dependency factories in
api/deps.py. Removed the channel methods + fields, deleted the
channel-specific stream.py endpoint, deleted require_channel_read/write.
Also deleted api/schemas/{channels,sessions}.py (hard dependency on the
removed enums; already fully dead after the Task 10 route deletions) and
api/schemas/messages.py (a TYPE_CHECKING-only import of the deleted
MessageTable; likewise already fully dead) + its dedicated test file.
Test updates: test_permissions.py -14 channel tests (matches the planned
count exactly), test_communications.py / test_communications_consumers.py
split to keep only notification-policy coverage, test_exceptions.py -9,
test_deps.py -4, plus the journal/stream/foundation-smoke fallout. Also
fixed a pre-existing (Task 7) broken assertion in
test_foundation_phase3_smoke.py that inspected a `say()` method already
removed from ContentActions.
Verified: full-suite collection (11961 tests, zero import errors) and a
complete test run (11567 passed, 394 skipped, 0 failed) in addition to
the targeted suites.
* migration: drop channels/groups/sessions/session_tasks/messages + enum types
alembic/versions/060_drop_messaging.py: drop_column journal_entries.
session_id (sidesteps hardcoding the FK constraint name — verified
empirically against a live migrated DB that it's actually
fk_journal_entries_session_id_sessions, but drop_column doesn't care
either way); drop_table in FK order (messages -> session_tasks ->
sessions -> groups -> channels); DROP TABLE IF EXISTS chunks_conversations
(runtime-provisioned, not alembic-managed, would otherwise orphan); DROP
TYPE IF EXISTS for messagetype/sessionstatus/sessionscope/channeltype
(messagetype's Python enum stays for ExtractedMessage, but the DB type
had zero live columns left once MessageTable was dropped in the prior
commit). downgrade() raises NotImplementedError — one-way removal.
Pruned scripts/reset_runtime_state.sql + .sh: removed the DELETE/COUNT
lines for messages/session_tasks/sessions/groups/channels and the
groups.active_session_id reset block.
Verified end-to-end against a scratch Postgres DB: full migration chain
001->060 applies cleanly, alembic heads shows a single head, all 6 dropped
tables + 4 enum types + the journal_entries.session_id column are
confirmed gone, journal_entries keeps only its journal_id/task_id FKs,
downgrade correctly raises NotImplementedError without corrupting DB
state, and the pruned reset_runtime_state.sql runs clean (no errors)
against a fully-migrated DB.
* refactor(api): remove channel/session/message routes + WS streams
Most of this task's file list was already forced through in earlier
commits (routes/{channels,groups,sessions,messages}.py + app.py/__init__.py
unmounting in the MessagingService-deletion commit; tasks.py's
session-links embed + GET /{id}/sessions + schemas/tasks.py's
TaskResponse.sessions in that same commit; deps.py's require_channel_read/
write + schemas/{channels,sessions}.py in the models/tables commit). This
closes out what was left:
- api/websocket.py: deleted the channel_stream + session_stream routes,
ConnectionManager's channel_connections/session_connections dicts,
connect_channel/connect_session, broadcast_to_channel/broadcast_to_session,
get_channel_subscriber_count, and their cleanup lines in disconnect().
Agent streams, notification streams, and the operator system stream are
untouched.
- api/websocket_bridge.py: deleted _handle_session_event +
_handle_message_event and their SESSION_CREATED/SESSION_CLOSED/
SESSION_TIMEOUT/MESSAGE_SENT subscriptions. The A2A live-view, rate-limit,
usage, agent-lifecycle, and notification bridges are untouched.
- api/schemas/websocket.py: removed NewMessageBroadcast, WSMessageNew,
WSMessageEdit, WSMessageDelete, WSSessionClosed — kept the WSMessage base
class (still subclassed by the kept WSAgentStream/WSNotification) plus
those two.
- api/schemas/groups.py: deleted (already fully orphaned since routes/
groups.py was removed; its GroupResponse/GroupDetailResponse had zero
consumers).
Updated the 5 websocket test files accordingly (removed the channel/
session-specific tests + fixed imports); test_websocket_bridge.py's
registration-coverage test dropped the SESSION_*/MESSAGE_SENT assertions.
Verified: full-suite collection (11943 tests, zero import errors) and a
complete test run (11549 passed, 394 skipped, 0 failed).
* docs: retire channels/sessions/messages from agent-facing docs + CLAUDE.md
Rewrites docs/rag (RAG-indexed) + docs/map + CLAUDE.md to reflect A2A (dm +
read_a2a) as primary agent comms; deletes the channel docs, splits messaging-tools
+ messaging-notification (renamed notification.md), swaps the WS worked example to
A2A_MESSAGE_SENT. _complete_map.md still needs regeneration (generated file).
* refactor(panel): remove Communications surface (channels/sessions)
Deletes the /communications routes, message components, task-detail Sessions tab,
use-channels + channel/session WS hooks, and the channels/sessions/messages/groups
api clients; prunes the Channel/Session/Message/Group types + mock data. (Auditor
live-feeds + dashboard.ts dead-route cleanup is a follow-up.)
* refactor(panel): drop auditor channel-feed + dead communication-metric route
* docs(map): regenerate _complete_map from updated slices
* fix(a2a): reduce get_unread_messages complexity below xenon C + stale comments
Extract the per-conversation unread-counter recompute into _reset_unread_counter
(the CI quality gate flagged get_unread_messages as rank C). Also drop the deleted
open_session from a content_actions comment and reword an evidence_repo docstring
that cited the removed messaging._notify_mentions.
---------
Co-authored-by: Renn F <rennf93@users.noreply.github.com>
39 KiB
RoboCo Slice Map — mcp-servers
Scope: roboco/mcp/ (every server file + schemas/ + utils.py). Repo root: /Users/renzof/Documents/GitHub/ZZZ/roboco-master/roboco.
Purpose
The roboco/mcp package is the agent-side MCP gateway: a set of FastMCP server processes that run inside each agent container and expose the RoboCo intent-verb / content-tool / RAG / docs / git-readonly / intake / secretary / web-research surfaces to the Claude Code or grok-CLI runtime as MCP tools. They are thin bridges — every tool either POSTs to the orchestrator's HTTP gateway (/api/v1/flow/*, /api/v1/do/*, /api/git/*, /optimal/*, /docs/*, /research/*, /api/secretary/*, /api/prompter/live/*) or, for the flow/do path, additionally forwards rejections to a per-container SDK loopback (ROBOCO_SDK_URL) that runs the per-verb circuit breaker. The orchestrator (not the MCP layer) is the authority for role scoping, state transitions, and git-side effects; the MCP layer only shapes calls, classifies rejections, and substitutes circuit_open envelopes when the breaker trips.
Files
| Path | Role | approx LOC |
|---|---|---|
roboco/mcp/__init__.py |
Package docstring only — deliberately import-free so python -m roboco.mcp.<name> does not pull sibling modules (esp. optimal_server's pgvector/ollama stack, ~6s startup). |
23 |
roboco/mcp/utils.py |
Shared HTTP helpers: _get_agent_headers, format_error_response, ApiResponse, ApiClient (async httpx wrapper). Used by optimal_server, docs_server, search_server. |
383 |
roboco/mcp/schemas/__init__.py |
Pydantic input models. After Phase-4 T9 deletions only WriteDocInput remains. |
35 |
roboco/mcp/flow_server.py |
roboco-flow MCP server — intent verbs (lifecycle). Manifest-scoped registration, role-scoped path /api/v1/flow/<route>/<verb>, per-verb circuit breaker + 404-route synthesis. |
1028 |
roboco/mcp/do_server.py |
roboco-do MCP server — content tools (commit, note, pitch, propose_roadmap, dm, notify, evidence, progress, playbook curation, pr_update). Manifest-scoped; fixed /api/v1/do/<verb> path; mirror circuit breaker. |
976 |
roboco/mcp/optimal_server.py |
roboco-optimal MCP server — RAG / KB / mentor / error / decision / standards / learnings / index-mgmt / proactive-context. Factory create_optimal_mcp_server(agent_id); calls /optimal/*. |
1102 |
roboco/mcp/docs_server.py |
roboco-docs MCP server — docs write/read/list/delete via /docs/*. Factory create_docs_mcp_server(agent_id). RAG-based dedup on write. |
251 |
roboco/mcp/git_readonly.py |
roboco-git-readonly MCP server — four read-only git views (status/log/diff/branch list) via /api/git/*. No breaker, no manifest. |
123 |
roboco/mcp/intake_server.py |
roboco-intake MCP server — grok intake path only: propose_draft / propose_batch POST directly to the prompter-live relay. |
215 |
roboco/mcp/secretary_server.py |
roboco-secretary MCP server — grok secretary path only: read_company_state / read_task / submit_directive, delegating to agent_sdk.secretary_driver. |
64 |
roboco/mcp/search_server.py |
roboco-search MCP server — web_search / web_fetch via /research/* (provider key stays server-side). Factory create_search_mcp_server(agent_id). |
130 |
(Excluded: __pycache__/, .DS_Store.)
Key Symbols
| Name | Kind | File:Line | Responsibility |
|---|---|---|---|
mcp (flow) |
FastMCP |
flow_server.py:252 |
Server instance roboco-flow; tools registered onto it at import time. |
mcp (do) |
FastMCP |
do_server.py:220 |
Server instance roboco-do. |
mcp (git-readonly) |
FastMCP |
git_readonly.py:31 |
Server instance roboco-git-readonly. |
mcp (intake) |
FastMCP |
intake_server.py:32 |
Server instance roboco-intake. |
mcp (secretary) |
FastMCP |
secretary_server.py:30 |
Server instance roboco-secretary. |
StrList |
type alias | flow_server.py:39 |
Annotated[list[str], BeforeValidator(coerce_str_list)] — tolerates Claude SDK's nested XML-ish tool-input shapes before MCP validation rejects. |
_CIRCUIT_REJECTION_KINDS |
frozenset | flow_server.py:66, do_server.py:50 |
The 4 breaker-counted kinds: tracing_gap, invalid_state, not_authorized, incomplete_input. |
_DICT_ERROR_CODE_MAP |
dict | flow_server.py:89, do_server.py:73 |
Exact code→kind map for known RobocoError codes (e.g. AUTHENTICATION_REQUIRED→not_authorized). Unknown codes fall through to a substring branch for forward-compat. Added in 536bbb64 to fix AUTHENTICATION_REQUIRED mis-routing (#161). |
_classify_dict_error_code |
func | flow_server.py:110, do_server.py:94 |
Map a dict-shaped error.code to a counted breaker kind: consults _DICT_ERROR_CODE_MAP first (exact), then substring fallback for unknown codes; NOT_FOUND → None. |
_remediate_for_kind |
func | flow_server.py:130, do_server.py:150 |
Synthesize a directed recovery hint string for each counted kind (not_found / incomplete_input / not_authorized / invalid_state). Used by _normalize_exception_envelope. |
_normalize_exception_envelope |
func | flow_server.py:164, do_server.py:180 |
Lift a dict-error exception-handler body or 422 detail list into Envelope wire format (string kind + message + remediate + missing). Returns None when payload is already a valid Envelope. Added in 0d714b6c (#232). |
_classify_rejection |
func | flow_server.py:216, do_server.py:114 |
Classify all 3 rejection shapes (string kind / dict error / 422 detail) → counted kind or None. Guards the dict in frozenset TypeError. |
_build_headers |
func | flow_server.py:256, do_server.py:224 |
Per-call headers: X-Agent-ID, X-Agent-Role, fresh X-Correlation-ID (UUID per MCP call). |
_post (flow) |
func | flow_server.py:272 |
POST to orchestrator; normalize exception bodies via _normalize_exception_envelope; synthesize invalid_state on bare 404 missing route; not_found on descriptive 404 detail; transport_error on non-JSON; forward rejection to breaker. |
_post (do) |
func | do_server.py:238 |
Mirror of flow _post for content tools. |
_verb_from_path |
func | flow_server.py:384, do_server.py:339 |
Extract verb name from path for breaker reporting. |
_record_and_check_circuit |
func | flow_server.py:394, do_server.py:349 |
Forward a rejection to SDK_URL/verb/attempted; if SDK says open, replace the payload with circuit_envelope (dict-copied, original nested as inner, task_id/correlation_id lifted to top level). Best-effort (fail-open). |
_ROLE_TO_ROUTE_PREFIX |
dict | flow_server.py:476 |
Maps product_owner/head_marketing → board route segment; every other role passes through unchanged. |
_role_path |
func | flow_server.py:483 |
Build /api/v1/flow/<route>/<verb> path. |
_TOOLS (flow) |
dict | flow_server.py:879 |
Verb name → Python impl map (27 verbs). pass/fail keys bridge the pass_review/fail_review IntentSpec names. |
_INTENT_TO_PUBLIC |
dict | flow_server.py:929 |
pass_review→pass, fail_review→fail — fixes the dogfood gap where QA tools were silently dropped. |
_load_manifest_flow_tools |
func | flow_server.py:935 |
Read /app/tool-manifest.json flow_tools; None if missing/unreadable. |
_register_tools (flow) |
func | flow_server.py:965 |
Raise RuntimeError if manifest missing and ROBOCO_ALLOW_FULL_TOOLSET not set; if set, registers full tool set as dev/test escape hatch. |
_REGISTERED_TOOLS (flow) |
var | flow_server.py:1024 |
Import-time registration side effect. |
_TOOLS (do) |
dict | do_server.py:863 |
Tool name → impl map (21 content tools, incl. propose_roadmap). |
_load_manifest_do_tools |
func | do_server.py:866 |
Read manifest do_tools list. |
_register_tools (do) |
func | do_server.py:896 |
Manifest-scoped registration; raise RuntimeError if manifest missing (unless ROBOCO_ALLOW_FULL_TOOLSET set). |
give_me_work … i_am_idle |
verb funcs | flow_server.py:491–625 |
Dev verbs. |
claim_review / pass_review / fail_review |
verb funcs | flow_server.py:628–653 |
QA verbs (registered as pass/fail). |
claim_pr_review / post_pr_review / claim_gate_review / pr_pass / pr_fail |
verb funcs | flow_server.py:656–876 |
PR-reviewer verbs (inbound + in-path gate). |
i_will_plan / delegate / submit_up / submit_root |
verb funcs | flow_server.py:759–856 |
PM coordination verbs. |
note |
verb func | do_server.py:428 |
Journal entry + handoff section writer (top-level done/next strings — the meltdown-#1 fix). |
commit / dm / notify / evidence |
verb funcs | do_server.py:423–598 |
Core content tools. |
propose_roadmap |
verb func | do_server.py:531 |
Product Owner (board-roadmap-only, _PRODUCT_OWNER_DO): propose a themed roadmap cycle (goal + 3-7 item drafts) exactly once per exploration task. |
draft_playbook / approve_playbook / reject_playbook / archive_playbook |
verb funcs | do_server.py:600–644 |
Playbook curation (delivery + Auditor). |
progress / notify_list / notify_get / notify_ack / pr_update / read_messages |
verb funcs | do_server.py:646–840 |
Wave-1 parity content tools. |
create_optimal_mcp_server |
factory | optimal_server.py:1068 |
Build roboco-optimal-{agent_id} server; registers 8 tool groups. |
roboco_kb_search / roboco_rag_query / roboco_kb_stats |
tools | optimal_server.py:70–213 |
Search / RAG / stats. |
roboco_ask_mentor |
tool | optimal_server.py:371 |
Primary conversational RAG tool (65s timeout). |
roboco_search_error / roboco_record_error_solution |
tools | optimal_server.py:439–535 |
Error-pattern memory. |
roboco_check_decision / roboco_record_decision |
tools | optimal_server.py:542–625 |
Decision memory (RecordDecisionInput pydantic model at L33). |
roboco_get_standards / roboco_validate_action / roboco_review_code |
tools | optimal_server.py:632–765 |
Standards + validation. |
roboco_record_learning / roboco_search_learnings |
tools | optimal_server.py:772–875 |
Learnings. |
roboco_clear_index / roboco_reindex_all / roboco_index_status |
tools | optimal_server.py:882–993 |
Index admin. |
roboco_get_proactive_context |
tool | optimal_server.py:1000 |
Stored-then-fresh proactive context for a task. |
normalize_index_types |
func | optimal_server.py:53 |
Map legacy docs alias → documentation before route's IndexType(...) conversion. |
create_docs_mcp_server |
factory | docs_server.py:160 |
Build roboco-docs-{agent_id} server (write/read/list/delete). |
WriteDocInput |
pydantic | schemas/__init__.py:12 |
Docs write input (task_id, filename, doc_type, title, content). |
roboco_git_status / roboco_git_log / roboco_git_diff / roboco_git_branch_list |
tools | git_readonly.py:45–119 |
Read-only git views via /api/git/*. |
propose_draft / propose_batch |
tools | intake_server.py:104–215 |
Grok intake: POST draft/batch to prompter-live relay; propose_batch drops malformed (no string title or name) entries via _normalize_batch_drafts and refuses empty batches. |
post_draft / post_batch / _post_event |
funcs | intake_server.py:41–102 |
Relay POST helpers (never raise; unit-testable with httpx.MockTransport). _post_event now captures relay response body under detail on non-success so the grok agent sees the real reason (0d714b6c). |
_draft_title |
func | intake_server.py:135 |
Extract a string title from a batch draft dict, accepting title or name key; returns None if neither is a string. |
_normalize_batch_drafts |
func | intake_server.py:146 |
Filter + normalize MegaTask batch drafts: drops title-less/name-less entries, normalizes name-only drafts onto title key. Returns (well_formed, dropped_count). |
read_company_state / read_task / submit_directive |
tools | secretary_server.py:33–60 |
Secretary CEO-authority tools; delegate to agent_sdk.secretary_driver. |
ApiClient |
class | utils.py:115 |
Async httpx client with agent headers, base URL settings.internal_api_url, get/post/put/patch/delete + *_or_error tuples. |
ApiResponse |
class | utils.py:82 |
Response wrapper (ok, status_code, json, text, is_status). |
_get_agent_headers |
func | utils.py:27 |
X-Agent-ID/X-Agent-Role/X-Agent-Team/X-Agent-Token (HMAC token from ROBOCO_AGENT_TOKEN). |
format_error_response |
func | utils.py:52 |
Wraps roboco.api.schemas.common.error_response. |
create_search_mcp_server |
factory | search_server.py:83 |
roboco-search-{agent_id} with web_search/web_fetch. |
Data Flow
Every server is launched as its own subprocess (uv run --no-sync python -m roboco.mcp.<name> [agent_id]) by the orchestrator's _generate_mcp_config (for flow/do/git-readonly/optimal/docs/search) or by the grok intake/secretary mains (for intake/secretary). At import time the flow/do servers read /app/tool-manifest.json (env ROBOCO_TOOL_MANIFEST_PATH) and register only the verbs/tools listed for the role — refusing to start if the manifest is missing, unless ROBOCO_ALLOW_FULL_TOOLSET is set (dev/test only; the all-tools fallback was the original bug that let PMs see dev verbs and 404). The optimal/docs/search/intake/secretary/git-readonly servers register their full tool surface unconditionally (role gating is server-side at the route).
When an agent calls a tool:
- flow/do —
_postbuilds headers (freshX-Correlation-IDper call), POSTs the JSON body to the orchestrator at/api/v1/flow/<route>/<verb>or/api/v1/do/<verb>. On 404 (a manifest-advertised verb with no matching route) it synthesizes aninvalid_stateEnvelope so the agent gets aremediatehint instead of a rawdetailbody. On non-JSON body it synthesizestransport_error. Otherwise the Envelope is surfaced as-is (success or rejection). - breaker path —
_classify_rejectiondetermines whether the envelope is a counted rejection (string kind / dicterror.code/ 422detaillist). If counted,_record_and_check_circuitPOSTs to the local SDK atROBOCO_SDK_URL/verb/attempted(2s timeout). If the SDK returnsopen=true, the original rejection is replaced by the SDK'scircuit_envelopebefore returning to the agent — stopping retry storms. SDK unreachable → fail-open (return original payload + log). - optimal/docs/search —
ApiClient(async httpx) calls the orchestrator's/optimal/*,/docs/*,/research/*routes withX-Agent-*headers; shapes the response into a tool-specific dict (status, results, hints). - git-readonly —
_getdoes a synchronous httpx GET to/api/git/*withX-Agent-ID/X-Agent-Role;raise_for_statuspropagates HTTP errors. - intake —
propose_draft/propose_batchPOST directly to/api/prompter/live/{session}/events(the prompter-live relay) because grok'sstreaming-jsonoutput does not surface tool-call events. Returns a human-readable string (not an Envelope). - secretary — the three tools delegate to
agent_sdk.secretary_driver._do_*helpers (shared with the Claude SDK path) andjson.dumpsthe result.
Mermaid
graph LR
subgraph AgentContainer["Agent container (per spawn)"]
CC["Claude Code / grok CLI runtime"]
FS["roboco-flow MCP"]
DS["roboco-do MCP"]
GR["roboco-git-readonly MCP"]
OP["roboco-optimal MCP"]
DOC["roboco-docs MCP (conditional)"]
SRCH["roboco-search MCP (conditional)"]
INT["roboco-intake MCP (grok only)"]
SEC["roboco-secretary MCP (grok only)"]
SDK["per-verb SDK loopback :9000"]
end
CC -->|MCP tool call| FS
CC -->|MCP tool call| DS
CC -->|MCP tool call| GR
CC -->|MCP tool call| OP
CC -->|MCP tool call| DOC
CC -->|MCP tool call| SRCH
CC -->|MCP tool call| INT
CC -->|MCP tool call| SEC
FS -->|POST /api/v1/flow/<role>/<verb>| ORC["Orchestrator HTTP gateway"]
DS -->|POST /api/v1/do/<verb>| ORC
GR -->|GET /api/git/*| ORC
OP -->|POST /optimal/*| ORC
DOC -->|POST /docs/*| ORC
SRCH -->|POST /research/*| ORC
INT -->|"POST /api/prompter/live/{s}/events"| ORC
SEC -->|POST /api/secretary/*| ORC
FS -.->|rejection → /verb/attempted| SDK
DS -.->|rejection → /verb/attempted| SDK
SDK -.->|open=true → circuit_envelope| FS
SDK -.->|open=true → circuit_envelope| DS
ORC -->|Envelope 2xx/4xx| FS
ORC -->|Envelope 2xx/4xx| DS
Logical Tree
roboco/mcp/
├── __init__.py # import-free package docstring (avoid sibling-load startup tax)
├── utils.py
│ ├── _get_agent_headers() # X-Agent-ID/Role/Team/Token (HMAC)
│ ├── format_error_response()
│ ├── ApiResponse # ok / status_code / json / text / is_status
│ └── ApiClient # async httpx; get/post/put/patch/delete + *_or_error
├── schemas/__init__.py
│ └── WriteDocInput # only survivor of Phase-4 T9 deletions
├── flow_server.py # roboco-flow (intent verbs)
│ ├── StrList # BeforeValidator(coerce_str_list) — SDK XML-ish input
│ ├── _CIRCUIT_REJECTION_KINDS / _DICT_ERROR_CODE_MAP / _classify_dict_error_code / _classify_rejection
│ ├── _remediate_for_kind / _normalize_exception_envelope
│ ├── _build_headers / _post / _verb_from_path / _record_and_check_circuit
│ ├── _ROLE_TO_ROUTE_PREFIX (PO/HM → board) / _role_path
│ ├── dev verbs: give_me_work, i_will_work_on, open_pr, i_am_done, i_am_blocked, unclaim, reassign, resume, sync_branch, i_am_idle
│ ├── QA verbs: claim_review, pass_review(→pass), fail_review(→fail)
│ ├── PR-reviewer verbs: claim_pr_review, post_pr_review, claim_gate_review, pr_pass, pr_fail
│ ├── Doc verbs: claim_doc_task, i_documented
│ ├── PM verbs: triage, triage_all, unblock, complete, escalate_up, i_will_plan, delegate, submit_up, submit_root
│ ├── Board/Main-PM: escalate_to_ceo
│ ├── _TOOLS / _INTENT_TO_PUBLIC
│ └── _load_manifest_flow_tools / _register_tools (fails loud if no manifest; ROBOCO_ALLOW_FULL_TOOLSET escape hatch)
├── do_server.py # roboco-do (content tools)
│ ├── mirror breaker machinery (_CIRCUIT_REJECTION_KINDS, _DICT_ERROR_CODE_MAP, _classify_*, _remediate_for_kind, _normalize_exception_envelope, _record_and_check_circuit)
│ ├── commit, note (handoff done/next), pitch, propose_roadmap, dm, notify, evidence
│ ├── progress, notify_list/get/ack, pr_update, read_messages
│ ├── draft_playbook, approve_playbook, reject_playbook, archive_playbook
│ └── _TOOLS / _load_manifest_do_tools / _register_tools (fails loud; ROBOCO_ALLOW_FULL_TOOLSET escape hatch)
├── optimal_server.py # roboco-optimal (RAG/KB)
│ ├── RecordDecisionInput, normalize_index_types (docs→documentation)
│ ├── _register_search_tools (kb_search, rag_query, kb_stats)
│ ├── _register_indexing_tools (index_code, index_docs)
│ ├── _register_utility_tools (tokens_estimate)
│ ├── _register_mentor_tools (ask_mentor)
│ ├── _register_error_tools (search_error, record_error_solution)
│ ├── _register_decision_tools (check_decision, record_decision)
│ ├── _register_standards_tools (get_standards, validate_action, review_code)
│ ├── _register_learning_tools (record_learning, search_learnings)
│ ├── _register_index_management_tools (clear_index, reindex_all, index_status)
│ ├── _register_proactive_tools (get_proactive_context)
│ └── create_optimal_mcp_server(agent_id)
├── docs_server.py # roboco-docs
│ ├── _handle_write/read/list/delete
│ └── create_docs_mcp_server(agent_id)
├── git_readonly.py # roboco-git-readonly (4 read-only tools)
├── intake_server.py # roboco-intake (grok only)
│ ├── _post_event / post_draft / post_batch
│ ├── _draft_title / _normalize_batch_drafts # title-or-name filter + name→title normalization
│ └── propose_draft / propose_batch (MegaTask)
├── secretary_server.py # roboco-secretary (grok only; delegates to secretary_driver)
└── search_server.py # roboco-search (web research, Board+PM)
├── _handle_search / _handle_fetch
└── create_search_mcp_server(agent_id)
Dependencies
Internal (roboco):
roboco.config.settings—internal_api_url(utils),research_enabled(orchestrator mount gate).roboco.agents_config—get_agent_role,get_agent_team(utils headers).roboco.api.schemas.common.error_response(utilsformat_error_response).roboco.foundation.policy.content.validators.coerce_str_list(flowStrList).roboco.agent_sdk.secretary_driver—_do_read_state/_do_read_task/_do_submit_directive(secretary server).roboco.mcp.schemas.WriteDocInput(docs server).roboco.mcp.utils.ApiClient/format_error_response(optimal, docs, search).
External:
mcp.server.fastmcp.FastMCP— MCP server framework (all servers).pydantic(BaseModel,Field,BeforeValidator,Annotated) — input validation.httpx— sync (flow/do/git-readonly) + async (utils ApiClient, intake) HTTP.structlog— logging (flow/do).fastapi.status— HTTP status constants (optimal).
Downstream consumers:
roboco.runtime.orchestrator._generate_mcp_config— mounts flow/do/git-readonly/optimal (always), docs (docs_roles), search (research_roles +research_enabled).roboco.agent_sdk.grok_intake_main/grok_secretary_main— mount intake/secretary for the grok path.roboco.runtime.spawn_manifest— writes/app/tool-manifest.json(theflow_tools/do_toolslists the flow/do servers read at import).
Entry Points
python -m roboco.mcp.flow_server—mcp.run()atflow_server.py:1028. Env required:ROBOCO_AGENT_ID,ROBOCO_AGENT_ROLE; readsROBOCO_ORCHESTRATOR_URL,ROBOCO_SDK_URL,ROBOCO_TOOL_MANIFEST_PATH.python -m roboco.mcp.do_server—mcp.run()atdo_server.py:954. Same env as flow.python -m roboco.mcp.git_readonly—mcp.run()atgit_readonly.py:123. Env:ROBOCO_AGENT_ID,ROBOCO_AGENT_ROLE,ROBOCO_ORCHESTRATOR_URL.python -m roboco.mcp.optimal_server <agent_id>—server.run()atoptimal_server.py:1102. Positionalagent_idarg.python -m roboco.mcp.docs_server <agent_id>—server.run()atdocs_server.py:251.python -m roboco.mcp.search_server <agent_id>—server.run()atsearch_server.py:130.python -m roboco.mcp.intake_server—mcp.run()atintake_server.py:215. Env:ROBOCO_API_URL,ROBOCO_PROMPTER_SESSION_ID. Mounted bygrok_intake_main, NOT by the orchestrator.python -m roboco.mcp.secretary_server—mcp.run()atsecretary_server.py:64. Env:ROBOCO_API_URL,ROBOCO_AGENT_ID,ROBOCO_AGENT_ROLE,ROBOCO_AGENT_TOKEN. Mounted bygrok_secretary_main, NOT by the orchestrator.
Invocation is one subprocess per agent container (the orchestrator writes roboco-mcp-{agent_id}.json into /app/mcp-configs and the runtime launches each mcpServers entry with uv run --no-sync pinned to /app/.venv).
Config Flags
Env vars read in this slice (all ROBOCO_*):
| Flag / env | Where | Purpose |
|---|---|---|
ROBOCO_AGENT_ID |
flow, do, git-readonly (required) | Agent identity for X-Agent-ID header + role-path. |
ROBOCO_AGENT_ROLE |
flow, do, git-readonly (required) | Role for X-Agent-Role + flow route prefix. |
ROBOCO_AGENT_TOKEN |
utils _get_agent_headers |
HMAC agent token injected by orchestrator at spawn; sent as X-Agent-Token. |
ROBOCO_ORCHESTRATOR_URL |
flow, do, git-readonly | Orchestrator base URL (default http://roboco-orchestrator:8000). |
ROBOCO_SDK_URL |
flow, do | Per-container SDK loopback for the breaker (default http://localhost:9000). |
ROBOCO_TOOL_MANIFEST_PATH |
flow, do | Path to the spawn manifest (default /app/tool-manifest.json). |
ROBOCO_API_URL |
intake, secretary | Orchestrator base URL for the grok-path servers. |
ROBOCO_PROMPTER_SESSION_ID |
intake | The live intake session id; without it propose_draft/propose_batch return a no-op string. |
ROBOCO_PROJECT_SLUG / ROBOCO_BRANCH |
set by orchestrator into mcp_env (consumed indirectly by /api/git/*) |
Git context. |
settings.research_enabled |
orchestrator mount gate for roboco-search (not read inside the slice) |
Web-research server armed only when true AND role in research_roles. |
settings.internal_api_url |
utils ApiClient.base_url |
Base URL for optimal/docs/search async calls. |
ROBOCO_ALLOW_FULL_TOOLSET |
flow _register_tools, do _register_tools |
Dev/test escape hatch: when set, a missing manifest registers the full tool set instead of raising RuntimeError. Never set in production — the full-toolset path was the original bug this policy replaced. |
No default-off feature flag is armed inside this slice; the only flag-gated server here is roboco-search (gated upstream by ROBOCO_RESEARCH_ENABLED in the orchestrator mount).
Gotchas
- Import-free
__init__.pyis load-bearing. Re-exporting server factories here would forceoptimal_server(pgvector/ollama stack, ~6s) to load on everypython -m roboco.mcp.<name>and time out the MCP init — symptom: "roboco-flow/do tools never register". - flow/do refuse to start without the manifest unless
ROBOCO_ALLOW_FULL_TOOLSETis set. A missing/app/tool-manifest.jsonraisesRuntimeErrorat import (production path). Previously the fallback registered all verbs, letting PMs call dev verbs at wrong URLs (404 storm). Local test runs without the bind mount can either setROBOCO_TOOL_MANIFEST_PATHto a real file or setROBOCO_ALLOW_FULL_TOOLSETto skip the hard-fail; the latter must never reach production containers. pass/failare Python keywords. The IntentSpec layer usespass_review/fail_review; the MCP layer exposes the public namespass/fail._INTENT_TO_PUBLICbridges the two. Forgetting this bridge silently drops QA tools from the palette (the dogfood bug that motivated it).- Dict-shaped
errorcrashes a naive breaker.error in frozensetraisesTypeError: unhashable type: 'dict'when FastAPI exception handlers returnerroras a dict._classify_rejectionusesisinstancechecks first — never adict in frozensetmembership test. - 404 handling has three cases (updated
536bbb64). (1) Bare default FastAPI 404 ({"detail": "Not Found"}) → missing route, synthesized asinvalid_statewith a wiring-gap remediate. (2) 404 carrying anerrorfield → surfaced as-is (proxy re-status edge case). (3) 404 with a descriptivedetailstring → surfaced asnot_foundwith a re-fetch remediate (#61). Previously only cases 1 and 2 existed and a descriptive 404 was mis-synthesized asinvalid_state. StrListis not just cosmetic. A barelist[str]annotation hard-rejects the Claude SDK's nested[[["…"]]]/[{item: {$text}}]tool-input shapes at MCP validation before the verb body runs — surfacing as a confusing1 validation error for i_will_planArguments…. TheBeforeValidatorflattens first.- Breaker is fail-open. SDK unreachable/slow/malformed → return the original rejection. The breaker is a safety net only; it must never break the gateway path.
_SDK_TIMEOUT=2.0is tight by design. note(scope='handoff')top-leveldone/nextare the load-bearing fields for PM resumption. Passing an emptysection={}used to crash the minimax PMs (done Field required→note circuit_open→ tracing gate blockeddelegate). The MCP signature now hasdone/nextas discrete string params. Do not passsection={}.propose_batchfilters and refuses empty batches. Drafts without a stringtitleORnameare dropped (via_draft_title); aname-only draft is normalized ontotitlebefore posting; if all are dropped it returns an error string instead of POSTing (would silently vanish on the panel side).droppedcount is sent to the relay. Previously onlytitlewas accepted —name-only drafts were silently dropped even if well-formed (536bbb64).- intake/secretary are NOT mounted by the orchestrator. They are mounted by
grok_intake_main/grok_secretary_mainfor the grok path only. The orchestrator's_generate_mcp_configonly knows flow/do/git-readonly/optimal/docs/search. optimal_serverpositionalagent_idis mandatory.python -m roboco.mcp.optimal_serverwith no arg prints usage and exits 1. Same for docs/search.normalize_index_typesmaps the legacydocsalias todocumentationbefore the route'sIndexType(...)conversion — without it agents passingindex_types=["docs"]get a 400.- git-readonly uses
raise_for_status. Unlike flow/do (which surface 4xx Envelopes), git-readonly propagates HTTP errors as exceptions. A non-200 from/api/git/*surfaces to the agent as a transport error, not an Envelope. X-Correlation-IDis minted per MCP call in flow/do (not per session). The orchestrator'sCorrelationIdMiddlewareaccepts it as the inbound id and binds structlog + audit row to it.
Drift from CLAUDE.md
CLAUDE.md "MCP servers running per agent container" table lists 5 servers (roboco-flow, roboco-do, roboco-git-readonly, roboco-optimal, roboco-docs). The actual roboco/mcp/ directory contains 8 server modules:
roboco/mcp/intake_server.py(roboco-intake) — omitted from the CLAUDE.md table. Mounted bygrok_intake_mainfor the grok intake path, not by_generate_mcp_config.roboco/mcp/secretary_server.py(roboco-secretary) — omitted from the CLAUDE.md table. Mounted bygrok_secretary_mainfor the grok secretary path.roboco/mcp/search_server.py(roboco-search) — omitted from the CLAUDE.md table. Mounted by_generate_mcp_config(orchestrator.py:2915) only whensettings.research_enabledAND role in(cell_pm, main_pm, product_owner, head_marketing).
CLAUDE.md "roboco-optimal" row says the server exposes roboco_ask_mentor, roboco_kb_search only. The actual optimal_server.py registers 18 tools (search/rag/stats/index_code/index_docs/tokens_estimate/ask_mentor/search_error/record_error_solution/check_decision/record_decision/get_standards/validate_action/review_code/record_learning/search_learnings/clear_index/reindex_all/index_status/get_proactive_context). Understatement, not contradiction.
CLAUDE.md "roboco-do" row lists commit, note, say, dm, evidence and (in the Agent Gateway section) draft_playbook for delivery roles + approve_playbook/reject_playbook/archive_playbook for the Auditor. The actual do_server.py do_tools registry also contains pitch, propose_roadmap (Product Owner only), progress, notify, notify_list, notify_get, notify_ack, pr_update, read_messages, read_a2a — none mentioned in CLAUDE.md (do_tools is 18 entries, not the 5 named). Understatement.
CLAUDE.md says the note/journal write "returns as soon as the entry is persisted; RAG indexing runs fire-and-forget." The MCP note tool itself is synchronous w.r.t. the orchestrator (a single POST); the fire-and-forget behavior is server-side, not visible in this slice — consistent, not drift.
No contradicted claims found in this slice; the drift is omission (3 servers, ~13 do-tools, ~16 optimal tools not listed).
Changes Since Baseline
Baseline: fd10cc862c2020b3f639cdb686d427b0198a2441. Commands:
git log --oneline fd10cc862c..HEAD -- roboco/mcp/
git diff --stat fd10cc862c..HEAD -- roboco/mcp/
Diff stat: do_server.py +140/-? , flow_server.py +172/-?, intake_server.py +20/-? (3 files, +298/-34).
Only one commit touched this slice since baseline:
15effce0— "Chore: 141 Gaps fill-in (#283)" (merged PR, 2026-06-29). IMPACT on this slice:flow_server.py: addedStrList(BeforeValidator(coerce_str_list)) so the Claude SDK's nested XML-ish tool-input shapes flatten before MCP validation; added_MISSING_ROUTE_STATUS404 handling that synthesizes aninvalid_stateEnvelope for manifest-registered verbs whose HTTP route is missing; added_classify_dict_error_code+_classify_rejectionso dict-shapederror(FastAPI exception handlers) and 422detail-list rejections count toward the per-verb circuit breaker (previously bypassed → unbounded retries); guarded againstTypeError: unhashable type: 'dict'.do_server.py: mirrored the same breaker machinery (the dogfood gap:note(scope='decision')had looped 8× returningincomplete_inputwith no breaker) + the same 404 missing-route synthesis.intake_server.py: docstring-only change —propose_draft/propose_batchtool descriptions now declare the per-cellproject_idonthe_work[]entries (MegaTask multi-cell fan-out). No logic change in intake.
No other commits in this slice since baseline.
Post-snapshot updates (since 2026-06-29):
536bbb64— "Chore/all/logical gaps sweep (#286)" (merged 2026-06-30). IMPACT on this slice:
flow_server.py+do_server.py: added_DICT_ERROR_CODE_MAP(exact code→kind map, replacing pure-substring classification; closes AUTHENTICATION_REQUIRED mis-routing #161); added descriptive-404 carve-out in_post(surfaced asnot_foundinstead ofinvalid_statefor real resource-not-found 404s, #61); the circuit_open substitution now dict-copies the SDK envelope and nests the original rejection asinner(#60);_register_toolsnow acceptsROBOCO_ALLOW_FULL_TOOLSETas a dev/test escape hatch instead of always raisingRuntimeError(#162).intake_server.py:propose_batchnow acceptsnameas a fallback fortitle(via new_draft_title/_normalize_batch_draftshelpers);name-only drafts are normalized ontotitlebefore posting rather than silently dropped (#163).docs_server.py:_handle_writenow surfaces acommit_status == "failed"outcome in the tool return string, telling the documenter to warn the cell PM when the doc could not be committed to the project repo (#34).
0d714b6c— "[chore] mcp-servers: normalize exception bodies to Envelope + lift task_id/correlation_id on circuit_open" (committed 2026-06-30). IMPACT on this slice:
flow_server.py+do_server.py: added_remediate_for_kindand_normalize_exception_envelope; the non-404 JSON path in_postnow normalizes dict-errorexception-handler bodies and 422detaillists into the Envelope wire format so agents getremediate/nextinstead of raw exception bodies (#232);_record_and_check_circuitliftstask_id/correlation_idfrom the original rejection onto the circuit_open envelope top level (#359).intake_server.py:_post_eventcaptures relay response body underdetailon non-success so the grok intake agent sees the real failure reason instead of an opaquehttp_422token (#57).
Regression Risks
| Title | File:Line | Claim | Severity |
|---|---|---|---|
| Breaker substitution could mask a real, fixable rejection | flow_server.py:394, do_server.py:349 |
Partially mitigated (536bbb64): the circuit_open envelope is now a dict-copy of the SDK's envelope with the original rejection nested as inner (preserving its kind/message/remediate). The agent sees circuit_open at the top level but the underlying rejection survives for ops debugging. The core risk remains: if the breaker trips on a mis-counted storm, the agent stops instead of retrying. |
medium |
| Dict-error classification substring fallback → mis-routing risk for unknown codes | flow_server.py:110, do_server.py:94 |
Partially mitigated (536bbb64): _classify_dict_error_code now consults _DICT_ERROR_CODE_MAP first (exact match for all known RobocoError codes, closing the AUTHENTICATION_REQUIRED mis-routing bug #161). Only codes NOT in the map fall through to the substring branch. A novel code that accidentally contains DENIED/AUTH/PERMISSION but is semantically different would still mis-route. Risk is now confined to future unknown codes only. |
low |
| 404 synthesis assumptions | flow_server.py:272, do_server.py:238 |
Partially mitigated (536bbb64 #61): a third 404 case was added: a 404 with a descriptive detail string (not the bare FastAPI default "Not Found") is now surfaced as not_found, not invalid_state. Residual risk: a future route that returns a bare 404 with no detail/error field for a real resource-not-found would still synthesize invalid_state. The two carve-outs (error field → as-is; descriptive detail → not_found) cover the known cases. |
low |
_register_tools raises at import if manifest missing — local dev breakage |
flow_server.py:965, do_server.py:896 |
Mitigated (536bbb64): ROBOCO_ALLOW_FULL_TOOLSET env var added as a dev/test escape hatch that bypasses the RuntimeError and registers the full tool set. Production behaviour is unchanged (ROBOCO_ALLOW_FULL_TOOLSET is not set in the orchestrator manifest). The mitigant must not leak into production containers. |
low |
propose_batch well-formed filter could silently drop intended drafts |
intake_server.py:146 |
Partially mitigated (536bbb64 #163): _draft_title now accepts name as a fallback for title, and _normalize_batch_drafts normalizes name-only drafts onto title before posting. Residual risk: a draft using a different key (e.g. label) is still dropped silently; the dropped count is sent but the CEO may not notice. |
low |
| Breaker SDK timeout (2s) may be too tight under load | flow_server.py:55, do_server.py:37 |
_SDK_TIMEOUT=2.0 for the loopback /verb/attempted POST. Under container CPU contention the SDK could exceed 2s and the breaker fails open (returns original payload) — re-introducing the unbounded-retry condition the breaker was added to stop. Fail-open is safe but defeats the protection. |
low |
Circuit-breaker forwarding does not pass task_id for content tools that lack one |
do_server.py:379 |
body.get("task_id") is sent to the SDK. Several do-tools (read_messages, notify_list) have no task_id — the breaker records task_id=None, so the per-verb (not per-task) breaker still works, but any future per-task breaker logic would mis-attribute these. |
low |
Health
The slice is coherent and well-defended: the flow/do servers share a near-identical, heavily-commented breaker/404-synthesis contract (with the duplication acknowledged in comments as a deliberate mirror), the manifest-gated registration is fail-loud and blocks the off-role-verb class, and the StrList / dict-error / 404 fixes added in 15effce0 close real observed retry-storm and validation-rejection loops. The main integrity concerns are (a) the duplicated breaker logic across flow/do is a drift hazard — a future change to _CIRCUIT_REJECTION_KINDS or classification must be applied in both files or the two servers diverge; (b) CLAUDE.md's server table is stale (3 servers + many tools unlisted), which could mislead a reader into thinking intake/secretary/search are not agent-facing MCP servers; (c) the breaker's fail-open posture is correct but means the protection is only as good as the SDK loopback staying responsive within 2s. No correctness bugs observed; the slice is fit for purpose.